Classification of myoelectric signal burst patterns using a dynamic neural network

The identification of physical signals is key to many signal processing applications. In the last decade, artificial neural networks have been shown to be a powerful tool for such pattern recognition tasks. Many signals are transient in nature, that is, they exist for only a limited duration in time. Moreover, much of the information in these transient bursts is conveyed by the dynamic evolution of the waveform in time the temporal structure of the signal. This is especially true of biological signals. Standard feedforward neural networks are not well-suited to capturing this temporal dimension. A neural network is described here that allows time to be represented implicitly within its structure, aiding its efficacy as a classifier of transient signals